Enterprise financial system abnormity identification method and system based on artificial intelligence

Abstract

The invention discloses an enterprise financial system abnormity identification method and system based on artificial intelligence. The method comprises the steps of data input, data processing, cost abnormity detection, transaction abnormity detection and enterprise financial system abnormity identification. The invention relates to the technical field of enterprise system exception prediction, in particular to an enterprise financial system exception identification method and system based on artificial intelligence, and the scheme performs enterprise financial system exception identification by combining two main exception detections of cost exception and transaction exception. Cost anomaly detection is carried out by adopting an integrated learning model combining an artificial neural network, a convolutional neural network and a deep residual network, and model parameters are optimized through gradient descent, so that the result availability and prediction efficiency of cost anomaly detection are improved; a hybrid model combining a deep neural network and a convolutional long-short-term memory network is adopted, the transaction feature processing capacity of the model is optimized, and the overall automation and reliability of the method are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of enterprise system anomaly prediction, and in particular to an enterprise financial system anomaly recognition method and system based on artificial intelligence. Background Art

[0002] The AI-based anomaly recognition method for enterprise financial systems is a technology that uses machine learning and deep learning algorithms to automatically detect and identify abnormal behaviors in enterprise financial data. It identifies potential fraud, errors or anomalies by analyzing financial statements, transaction records, and cost data, helping enterprises to promptly discover and deal with financial risks and improve the accuracy and efficiency of financial management.

[0003] However, in the existing methods for identifying anomalies in enterprise financial systems, there are technical problems that the tasks of enterprise financial systems are complex, the data styles are complex and the dimensions are deep, while traditional enterprise financial system anomaly identification often focuses on a single detection or monitoring task, resulting in insufficient analysis of anomaly identification; in the existing cost anomaly detection, there is a technical problem that cost anomaly detection in traditional methods uses a single model for simple machine learning data prediction, and the prediction of a single model is relatively independent, making it difficult to connect and integrate with subsequent transaction anomaly detection, and the accuracy and efficiency of single model prediction detection need to be improved; in the existing transaction anomaly detection, there is a technical problem that traditional transaction anomaly detection methods generally use simple models for prediction, and the prediction of a single model has similar problems as cost anomaly detection, and both are difficult to expand and integrate. The data dimension of transaction anomaly detection is more complex than that of cost data, so there is an urgent need to design an advanced hybrid model method suitable for integration. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an enterprise financial system anomaly identification method and system based on artificial intelligence. In view of the technical problem that in the existing enterprise financial system anomaly identification method, the enterprise financial system tasks are complicated, the data style is complex and the dimension is deep, and the traditional enterprise financial system anomaly identification often focuses on a single detection or monitoring task, resulting in insufficient analysis of anomaly identification, this solution creatively adopts the combination of cost anomaly and transaction anomaly two main anomaly detection to identify anomalies in the enterprise financial system. Through the combination and expansion of tasks, the depth and breadth of anomaly identification are improved, and the overall usability and versatility of enterprise financial system anomaly identification are improved; in view of the technical problem that in the existing cost anomaly detection, cost anomaly detection uses a single model to perform simple machine learning data prediction in traditional methods, and the prediction of the single model is relatively independent, and it is difficult to connect and integrate with subsequent transaction anomaly detection, and the accuracy and efficiency of single model prediction detection need to be improved ... An integrated learning model combining artificial neural networks, convolutional neural networks and deep residual networks is used for cost anomaly detection. By deeply mining and learning nonlinear data and local features, optimizing model parameters through gradient descent, and finally performing integrated learning of model output, the availability, accuracy and prediction efficiency of cost anomaly detection results are improved. In the existing transaction anomaly detection, traditional transaction anomaly detection methods generally use simple models for prediction, and the prediction of a single model has similar problems as cost anomaly detection, and both are difficult to expand and integrate. The data dimension of transaction anomaly detection is more complex than that of cost data. Therefore, it is urgent to design a technical problem of an advanced hybrid model method suitable for integration. This scheme creatively adopts a hybrid model combining deep neural networks and convolutional long short-term memory networks for transaction anomaly detection, and by selecting convolutional networks and long short-term memory networks to replace the hidden layers of deep neural networks, the model's processing ability for transaction features is optimized, and the prediction depth of the model and the overall automation and reliability of the method are improved.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides an enterprise financial system anomaly recognition method based on artificial intelligence, the method comprising the following steps:

[0006] Step S1: data input;

[0007] Step S2: data processing;

[0008] Step S3: cost anomaly detection;

[0009] Step S4: transaction anomaly detection;

[0010] Step S5: Identify abnormalities in the enterprise financial system.

[0011] Further, in step S1, the data input is used to receive the original data of the enterprise's financial related data, specifically by connecting to the enterprise's financial system database, collecting the enterprise's historical data, and receiving the real-time financial data stream by building a structure to obtain the original data for abnormal identification of the financial system;

[0012] The financial system anomaly identification original data specifically includes financial statement data, transaction data, cost data and associated data.

[0013] Furthermore, in step S2, the data processing is used to optimize the input data, specifically, to sequentially perform data cleaning, data preprocessing and feature engineering design on the original data of financial system anomaly identification to obtain financial system anomaly identification optimized data and financial system anomaly identification feature data.

[0014] Further, in step S3, the cost anomaly detection is used to detect whether there is abnormal cost data in the enterprise cost data, specifically, based on the financial system anomaly identification optimization data and the financial system anomaly identification feature data, an integrated learning model combining an artificial neural network, a convolutional neural network and a deep residual network is used to perform cost anomaly detection to obtain financial system cost anomaly feedback data;

[0015] The integrated learning model combining artificial neural network, convolutional neural network and deep residual network specifically includes artificial neural subnet, convolutional neural subnet, deep residual subnet and gradient descent operator;

[0016] The artificial neural subnet is used to optimize the processing of nonlinear feature relationships;

[0017] The convolutional neural subnet is used to extract local feature information of the data;

[0018] The deep residual subnet is used to optimize the model depth and learn more complex feature data;

[0019] The gradient descent operator is used to adjust and optimize model parameters and model performance;

[0020] The step of using an integrated learning model combining an artificial neural network, a convolutional neural network and a deep residual network to detect cost anomalies and obtain the cost anomaly feedback data of the financial system includes:

[0021] Step S31: constructing an artificial neural subnet, specifically by constructing a multilayer perceptron and performing nonlinear activation to obtain an artificial neural subnet prediction output;

[0022] Step S32: construct a convolutional neural subnet, specifically construct a convolutional neural subnet including a convolutional layer and a maximum pooling layer, obtain a convolutional subnet prediction output, and connect the convolutional subnet prediction output to the deep residual subnet;

[0023] Step S33: construct a deep residual subnet, specifically construct a residual block for residual learning, and activate it through a ReLU activation function to obtain a deep residual subnet prediction output;

[0024] Step S34: constructing a gradient descent operator, specifically, updating the learning rate and loss function of the model through iterative calculation, and iteratively optimizing the model parameters of the artificial neural subnet, the convolutional neural subnet, and the deep residual subnet;

[0025] Step S35: subnetwork model integration, specifically integrating model output values ​​by weighted summation to obtain integrated model prediction data;

[0026] Step S36: Cost anomaly detection model training, specifically, by constructing the artificial neural subnet, the convolutional neural subnet, the deep residual subnet, the gradient descent operator and the subnet model integration, model training is performed to obtain the cost anomaly detection model Model CD ;

[0027] Step S37: Cost anomaly detection, specifically using the cost anomaly detection model Model CD , based on the financial system anomaly identification optimization data and the financial system anomaly identification feature data, cost anomaly detection is performed to obtain financial system cost anomaly feedback data.

[0028] Further, in step S4, the transaction anomaly detection is used to automatically detect potential abnormal risks in the transaction data, specifically, based on the financial system cost anomaly feedback data and the financial system anomaly identification optimization data, a hybrid model combining a deep neural network and a convolutional long short-term memory network is used to perform transaction anomaly detection to obtain transaction anomaly detection data;

[0029] The hybrid model combining the deep neural network and the convolutional long short-term memory network specifically includes a convolutional layer, a deep neural layer and a long short-term memory layer;

[0030] The convolutional layer is used to extract local spatial features in the input data and serves as a hidden layer of the deep neural layer;

[0031] The deep neural layer is used to construct the basic input-output structure of the hybrid model;

[0032] The long short-term memory layer is used to extract the temporal features in the input data and serves as a hidden layer of the deep neural layer;

[0033] The step of using a hybrid model combining a deep neural network and a convolutional long short-term memory network to perform transaction anomaly detection and obtain transaction anomaly detection data includes:

[0034] Step S41: constructing a convolution layer, specifically constructing a basic convolution layer including convolution weights and bias items, and using a nonlinear activation function ReLU to activate, obtain convolution feature output, and constructing the convolution layer as the first hidden layer of the deep neural layer;

[0035] Step S42: construct a deep neural layer, specifically construct a deep neural layer including an input layer, a first hidden layer, a second hidden layer and an output layer, and use the ReLU activation function to activate the two hidden layers to obtain the deep neural layer prediction output

[0036] Step S43: constructing a long short-term memory layer, specifically constructing a long short-term memory layer including a forget gate, an input gate, an output gate, a candidate cell state and a hidden state, and constructing the long short-term memory layer as a second hidden layer of the deep neural layer;

[0037] Step S44: Model hybrid connection, specifically, replacing the first hidden layer of the deep neural layer with a convolutional layer, and replacing the second hidden layer with a long short-term memory layer, performing model hybrid connection, and constructing a hybrid model combining a deep neural network and a convolutional long short-term memory network;

[0038] Step S45: Transaction anomaly detection model training, specifically, training the transaction anomaly detection model by constructing the convolution layer, the deep neural layer, the long short-term memory layer and the model hybrid connection to obtain the transaction anomaly detection model Model ED ;

[0039] Step S46: Transaction anomaly detection, specifically using the transaction anomaly detection model Model ED , based on the financial system cost anomaly feedback data and the financial system anomaly identification optimization data, transaction legacy detection is performed to obtain transaction anomaly detection data.

[0040] Further, in step S5, the enterprise financial system anomaly identification is used to integrate the anomaly detection results of costs and transactions to identify the overall anomaly of the enterprise financial system, specifically by adopting a weighted average method to fuse the transaction anomaly detection data and the financial system cost anomaly feedback data, and by assigning a weight to each detection result, calculating a comprehensive anomaly score, and setting an anomaly detection threshold based on the comprehensive anomaly score, and in the real-time monitoring of the enterprise financial system, when cost data or transaction data exceeding the anomaly threshold is monitored, an anomaly detection report is generated to obtain anomaly identification reference data for the enterprise financial system;

[0041] The enterprise financial system anomaly identification reference data specifically includes anomaly descriptions and anomaly correction suggestions.

[0042] The present invention provides an enterprise financial system anomaly identification system based on artificial intelligence, comprising a data input module, a data processing module, a cost anomaly detection module, a transaction anomaly detection module and an enterprise financial system anomaly identification module;

[0043] The data input module is used for data input, and obtains the original data of abnormality identification of the financial system through data input, and sends the original data of abnormality identification of the financial system to the data processing module;

[0044] The data processing module is used for data processing, and obtains financial system anomaly identification optimization data and financial system anomaly identification feature data through data processing, and sends the financial system anomaly identification optimization data and financial system anomaly identification feature data to the cost anomaly detection module and the transaction anomaly detection module;

[0045] The cost anomaly detection module is used for cost anomaly detection, obtains financial system cost anomaly feedback data through cost anomaly detection, and sends the financial system cost anomaly feedback data to the transaction anomaly detection module and the enterprise financial system anomaly identification module;

[0046] The transaction anomaly detection module is used for transaction anomaly detection, obtains transaction anomaly detection data through transaction anomaly detection, and sends the transaction anomaly detection data to the enterprise financial system anomaly identification module;

[0047] The enterprise financial system anomaly identification module is used for enterprise financial system anomaly identification, and obtains enterprise financial system anomaly identification reference data through enterprise financial system anomaly identification.

[0048] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0049] (1) In view of the technical problems that the existing methods for identifying anomalies in enterprise financial systems have complex tasks, complex data patterns and deep dimensions, and traditional anomaly identification in enterprise financial systems often focuses on a single detection or monitoring task, resulting in insufficient analysis of anomaly identification, this solution creatively adopts the combination of cost anomaly and transaction anomaly detection to identify anomalies in enterprise financial systems. By combining and expanding tasks, the depth and breadth of anomaly identification are improved, and the overall usability and versatility of anomaly identification in enterprise financial systems are improved;

[0050] (2) In the existing cost anomaly detection, there are technical problems that the traditional method of cost anomaly detection uses a single model to perform simple machine learning data prediction, and the prediction of a single model is relatively independent and difficult to connect and integrate with subsequent transaction anomaly detection. In addition, the accuracy and efficiency of single model prediction detection need to be improved. This solution creatively uses an integrated learning model that combines artificial neural networks, convolutional neural networks, and deep residual networks to perform cost anomaly detection. By deeply mining and learning nonlinear data and local features, optimizing model parameters through gradient descent, and finally performing integrated learning of model output, the availability, accuracy, and prediction efficiency of cost anomaly detection results are improved.

[0051] (3) In the existing transaction anomaly detection, the traditional transaction anomaly detection method generally uses a simple model for prediction. The prediction of a single model has similar problems to cost anomaly detection, and both are difficult to expand and integrate. The data dimension of transaction anomaly detection is more complex than that of cost data. Therefore, it is urgent to design an advanced hybrid model method suitable for integration. This scheme creatively adopts a hybrid model combining deep neural networks and convolutional long short-term memory networks for transaction anomaly detection. By selecting convolutional networks and long short-term memory networks to replace the hidden layers of deep neural networks, the model's processing ability for transaction features is optimized, and the prediction depth of the model and the overall automation and reliability of the method are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 A flowchart of an artificial intelligence-based enterprise financial system anomaly identification method provided by the present invention;

[0053] Figure 2 A schematic diagram of an enterprise financial system anomaly recognition system based on artificial intelligence provided by the present invention;

[0054] Figure 3 This is a schematic diagram of the data processing flow in step S2;

[0055] Figure 4 This is a flow chart of cost anomaly detection in step S3;

[0056] Figure 5 This is a flow chart of transaction anomaly detection in step S4.

[0057] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0058] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0059] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0060] Example 1, see Figure 1 The present invention provides an artificial intelligence-based enterprise financial system anomaly recognition method, which includes the following steps:

[0061] Step S1: data input;

[0062] Step S2: data processing;

[0063] Step S3: cost anomaly detection;

[0064] Step S4: transaction anomaly detection;

[0065] Step S5: Identify abnormalities in the enterprise financial system.

[0066] By performing the above operations, in order to address the technical problem that in the existing enterprise financial system anomaly identification methods, the enterprise financial system tasks are complicated, the data styles are complex and the dimensions are deep, and the traditional enterprise financial system anomaly identification often focuses on a single detection or monitoring task, resulting in insufficient analysis of anomaly identification, this solution creatively adopts the combination of cost anomaly and transaction anomaly detection to identify anomalies in the enterprise financial system. Through the combination and expansion of tasks, the depth and breadth of anomaly identification are improved, and the overall usability and versatility of enterprise financial system anomaly identification are improved.

[0067] Example 2, see Figure 1 and Figure 2 ,In step S1, the data input is used to receive the original data of enterprise financial related data, specifically by connecting to the enterprise financial system database, collecting the enterprise historical data, and receiving the real-time financial data stream by building a structure to obtain the original data for abnormal identification of the financial system;

[0068] The financial system abnormality identification original data specifically includes financial statement data, transaction data, cost data and related data;

[0069] The financial statement data specifically includes a balance sheet, a profit sheet and a cash flow sheet;

[0070] The transaction data specifically includes sales records, purchase records, expense records and revenue records;

[0071] The cost data specifically includes raw material costs, labor costs and operating costs;

[0072] The associated data specifically includes tax record data and bank reconciliation data.

[0073] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data processing is used to optimize the input data, specifically, data cleaning, data preprocessing and feature engineering design are performed on the original data of financial system abnormality identification in sequence to obtain financial system abnormality identification optimization data and financial system abnormality identification feature data, which specifically includes the following steps:

[0074] Step S21: data cleaning, specifically, performing data deduplication, missing value filling and data format standardization operations on the original data of the financial system abnormality identification to obtain financial system cleansing data;

[0075] Step S22: data preprocessing, specifically performing data normalization and differential time series data processing operations on the financial system cleansing data to obtain financial system anomaly identification optimization data;

[0076] Step S23: feature engineering design, specifically, based on the financial system anomaly identification optimization data, a diversified quantitative feature extraction combining hierarchical clustering and principal component analysis is used to obtain financial system anomaly identification feature data;

[0077] The diversified quantitative feature extraction using hierarchical clustering and principal component analysis specifically includes the following steps:

[0078] Step S231: clustering variables, specifically, clustering data variables in the financial system anomaly identification optimization data with high correlation using a hierarchical clustering algorithm to obtain a financial system cluster variable group;

[0079] Step S232: feature analysis, specifically, using principal component analysis to perform principal component analysis on the variables in the financial system clustering variable group, extracting the first principal component of each variable as a data feature, and obtaining financial system anomaly identification feature data.

[0080] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the cost anomaly detection is used to detect whether there is abnormal cost data in the enterprise cost data. Specifically, based on the financial system anomaly identification optimization data and the financial system anomaly identification feature data, an integrated learning model combining artificial neural network, convolutional neural network and deep residual network is used to perform cost anomaly detection to obtain financial system cost anomaly feedback data;

[0081] The integrated learning model combining artificial neural network, convolutional neural network and deep residual network specifically includes artificial neural subnet, convolutional neural subnet, deep residual subnet and gradient descent operator;

[0082] The artificial neural subnet is used to optimize the processing of nonlinear feature relationships;

[0083] The convolutional neural subnet is used to extract local feature information of the data;

[0084] The deep residual subnet is used to optimize the model depth and learn more complex feature data;

[0085] The gradient descent operator is used to adjust and optimize model parameters and model performance;

[0086] The step of using an integrated learning model combining an artificial neural network, a convolutional neural network and a deep residual network to detect cost anomalies and obtain the cost anomaly feedback data of the financial system includes:

[0087] Step S31: constructing an artificial neural subnet, specifically by constructing a multilayer perceptron and performing nonlinear activation to obtain an artificial neural subnet prediction output;

[0088] The calculation formula of the artificial neural subnet prediction output is:

[0089] ;

[0090] Where A is the predicted output of the artificial neural subnet, sig(·) is the S-type activation function, L is the level index of the artificial neural subnet, and w L is the weight matrix of the artificial neural subnetwork at layer L, a L-1 is the output of the previous L-1 layer, b L is the bias term of the artificial neural subnetwork at layer L;

[0091] Step S32: construct a convolutional neural subnet, specifically construct a convolutional neural subnet including a convolutional layer and a maximum pooling layer, obtain a convolutional subnet prediction output, and connect the convolutional subnet prediction output to the deep residual subnet;

[0092] The calculation formula for the convolutional subnet prediction output is:

[0093] C = softmax(W k *x+b k );

[0094] Where C is the predicted output of the convolutional subnet, softmax(·) is the classifier function, k is the convolution kernel index, and W k is the weight parameter of the kth convolution kernel, x is the data input, which is used to represent the financial system anomaly recognition optimization data and the financial system anomaly recognition feature data, b k is the bias term of the kth convolution kernel;

[0095] Step S33: construct a deep residual subnet, specifically construct a residual block for residual learning, and activate it through a ReLU activation function to obtain a deep residual subnet prediction output;

[0096] The calculation formula of the deep residual subnet prediction output is:

[0097] R=F(x,W i )+x;

[0098] Where R is the prediction output of the deep residual subnet, F(x,W i ) is the residual block mapping function, which is used to represent the convolutional subnet prediction output, W i is the weight matrix of the i-th residual block, where i is the residual block index and x is the data input;

[0099] Step S34: construct a gradient descent operator, specifically, update the learning rate and loss function of the model through iterative calculation, perform iterative optimization of the model parameters of the artificial neural subnet, convolutional neural subnet and deep residual subnet, and obtain the optimal model parameters. The calculation formula is:

[0100] ;

[0101] In the formula, is the parameter value of the model in the t+1th iteration, is the parameter value of the model in the tth iteration, t is the iteration number index, is the learning rate parameter, is the model loss function L with respect to the model parameters The gradient calculation value of is used to represent the change value of the loss function at the current model parameters;

[0102] Step S35: subnetwork model integration, specifically integrating model output values ​​by weighted summation to obtain integrated model prediction data;

[0103] The specific calculation formula for the integrated model prediction data is:

[0104] ;

[0105] In the formula, Z is the prediction data of the integrated model, A is the prediction output of the artificial neural subnet, R is the prediction output of the deep residual subnet, and R is the prediction output of the deep residual subnet;

[0106] Step S36: Cost anomaly detection model training, specifically, by constructing the artificial neural subnet, the convolutional neural subnet, the deep residual subnet, the gradient descent operator and the subnet model integration, model training is performed to obtain the cost anomaly detection model Model CD ;

[0107] Step S37: Cost anomaly detection, specifically using the cost anomaly detection model Model CD , based on the financial system anomaly identification optimization data and the financial system anomaly identification feature data, cost anomaly detection is performed to obtain financial system cost anomaly feedback data.

[0108] By performing the above operations, in the existing cost anomaly detection, there is a technical problem that the traditional method of cost anomaly detection uses a single model to perform simple machine learning data prediction, and the prediction of the single model is relatively independent, which is difficult to connect and integrate with the subsequent transaction anomaly detection, and the accuracy and efficiency of the single model prediction detection need to be improved. This solution creatively uses an integrated learning model that combines artificial neural networks, convolutional neural networks and deep residual networks to perform cost anomaly detection. By deeply mining and learning nonlinear data and local features, optimizing model parameters through gradient descent, and finally performing integrated learning of model output, the availability, accuracy and prediction efficiency of the cost anomaly detection results are improved.

[0109] Example 5, see Figure 1 , Figure 2 and Figure 5 This embodiment is based on the above embodiment. In step S4, the transaction anomaly detection is used to automatically detect potential abnormal risks in transaction data. Specifically, based on the financial system cost anomaly feedback data and the financial system anomaly identification optimization data, a hybrid model combining a deep neural network and a convolutional long short-term memory network is used to perform transaction anomaly detection to obtain transaction anomaly detection data.

[0110] The hybrid model combining the deep neural network and the convolutional long short-term memory network specifically includes a convolutional layer, a deep neural layer and a long short-term memory layer;

[0111] The convolutional layer is used to extract local spatial features in the input data and serves as a hidden layer of the deep neural layer;

[0112] The deep neural layer is used to construct the basic input-output structure of the hybrid model;

[0113] The long short-term memory layer is used to extract the temporal features in the input data and serves as a hidden layer of the deep neural layer;

[0114] The step of using a hybrid model combining a deep neural network and a convolutional long short-term memory network to perform transaction anomaly detection and obtain transaction anomaly detection data includes:

[0115] Step S41: constructing a convolution layer, specifically constructing a basic convolution layer including convolution weights and bias items, and using a nonlinear activation function ReLU to activate, obtain convolution feature output, and constructing the convolution layer as the first hidden layer of the deep neural layer;

[0116] Step S42: construct a deep neural layer, specifically construct a deep neural layer including an input layer, a first hidden layer, a second hidden layer and an output layer, and use the ReLU activation function to activate the two hidden layers to obtain the deep neural layer prediction output

[0117] Step S43: constructing a long short-term memory layer, specifically constructing a long short-term memory layer including a forget gate, an input gate, an output gate, a candidate cell state and a hidden state, and constructing the long short-term memory layer as a second hidden layer of the deep neural layer;

[0118] Step S44: Model hybrid connection, specifically, replacing the first hidden layer of the deep neural layer with a convolutional layer, and replacing the second hidden layer with a long short-term memory layer, performing model hybrid connection, and constructing a hybrid model combining a deep neural network and a convolutional long short-term memory network;

[0119] Step S45: Transaction anomaly detection model training, specifically, training the transaction anomaly detection model by constructing the convolution layer, the deep neural layer, the long short-term memory layer and the model hybrid connection to obtain the transaction anomaly detection model Model ED ;

[0120] Step S46: Transaction anomaly detection, specifically using the transaction anomaly detection model Model ED , based on the financial system cost anomaly feedback data and the financial system anomaly identification optimization data, transaction legacy detection is performed to obtain transaction anomaly detection data.

[0121] By performing the above operations, in the existing transaction anomaly detection, the traditional transaction anomaly detection method generally uses a simple model for prediction, and the prediction of a single model has similar problems as the cost anomaly detection, which are difficult to expand and integrate. The data dimension of transaction anomaly detection is more complex than the cost data. Therefore, it is urgent to design a technical problem of an advanced hybrid model method suitable for integration. This scheme creatively adopts a hybrid model combining deep neural networks and convolutional long short-term memory networks for transaction anomaly detection, and by selecting convolutional networks and long short-term memory networks to replace the hidden layers of deep neural networks, the model's processing ability for transaction features is optimized, and the prediction depth of the model and the overall automation and reliability of the method are improved.

[0122] Embodiment 7, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment. In step S5, the enterprise financial system anomaly identification is used to integrate the anomaly detection results of costs and transactions to identify the overall anomaly of the enterprise financial system. Specifically, the transaction anomaly detection data and the financial system cost anomaly feedback data are merged by the weighted average method, and a comprehensive anomaly score is calculated by assigning a weight to each detection result. The anomaly detection threshold is set according to the comprehensive anomaly score, and in the real-time monitoring of the enterprise financial system, when cost data or transaction data exceeding the anomaly threshold is monitored, an anomaly detection report is generated to obtain the enterprise financial system anomaly identification reference data;

[0123] The enterprise financial system anomaly identification reference data specifically includes anomaly descriptions and anomaly correction suggestions.

[0124] Embodiment 8, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides an enterprise financial system anomaly identification system based on artificial intelligence, including a data input module, a data processing module, a cost anomaly detection module, a transaction anomaly detection module and an enterprise financial system anomaly identification module;

[0125] The data input module is used for data input, and obtains the original data of abnormality identification of the financial system through data input, and sends the original data of abnormality identification of the financial system to the data processing module;

[0126] The data processing module is used for data processing, and obtains financial system anomaly identification optimization data and financial system anomaly identification feature data through data processing, and sends the financial system anomaly identification optimization data and financial system anomaly identification feature data to the cost anomaly detection module and the transaction anomaly detection module;

[0127] The cost anomaly detection module is used for cost anomaly detection, obtains financial system cost anomaly feedback data through cost anomaly detection, and sends the financial system cost anomaly feedback data to the transaction anomaly detection module and the enterprise financial system anomaly identification module;

[0128] The transaction anomaly detection module is used for transaction anomaly detection, obtains transaction anomaly detection data through transaction anomaly detection, and sends the transaction anomaly detection data to the enterprise financial system anomaly identification module;

[0129] The enterprise financial system anomaly identification module is used for enterprise financial system anomaly identification, and obtains enterprise financial system anomaly identification reference data through enterprise financial system anomaly identification.

[0130] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0131] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0132] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for identifying abnormalities in enterprise financial systems based on artificial intelligence, characterized in that: The method comprises the following steps: Step S1: Data input, obtaining the original data of abnormal identification of the financial system; Step S2: Data processing to obtain financial system abnormality identification optimization data and financial system abnormality identification feature data; Step S3: cost anomaly detection, which is used to detect whether there is abnormal cost data in the enterprise cost data through multi-model integrated learning, specifically, based on the financial system anomaly identification optimization data and the financial system anomaly identification feature data, an integrated learning model combining artificial neural network, convolutional neural network and deep residual network is used to perform cost anomaly detection to obtain financial system cost anomaly feedback data; Step S4: transaction anomaly detection, which is used to automatically detect potential abnormal risks in transaction data through a hybrid model, specifically, based on the financial system cost anomaly feedback data and the financial system anomaly identification optimization data, a hybrid model combining a deep neural network and a convolutional long short-term memory network is used to perform transaction anomaly detection to obtain transaction anomaly detection data; Step S5: Enterprise financial system anomaly identification, which is used to integrate the anomaly detection results of costs and transactions to obtain enterprise financial system anomaly identification reference data.

2. The method for identifying abnormalities in an enterprise financial system based on artificial intelligence according to claim 1, characterized in that: In step S3, the integrated learning model combining the artificial neural network, the convolutional neural network and the deep residual network specifically includes an artificial neural subnet, a convolutional neural subnet, a deep residual subnet and a gradient descent operator; The artificial neural subnet is used to optimize the processing of nonlinear feature relationships; The convolutional neural subnet is used to extract local feature information of the data; The deep residual subnet is used to optimize the model depth and learn more complex feature data; The gradient descent operator is used to adjust and optimize model parameters and model performance.

3. The method for identifying abnormalities in an enterprise financial system based on artificial intelligence according to claim 2 is characterized in that: In step S3, the step of using an integrated learning model combining an artificial neural network, a convolutional neural network and a deep residual network to perform cost anomaly detection to obtain financial system cost anomaly feedback data includes: Step S31: constructing an artificial neural subnet, specifically by constructing a multilayer perceptron and performing nonlinear activation to obtain an artificial neural subnet prediction output; Step S32: construct a convolutional neural subnet, specifically construct a convolutional neural subnet including a convolutional layer and a maximum pooling layer, obtain a convolutional subnet prediction output, and connect the convolutional subnet prediction output to the deep residual subnet; Step S33: construct a deep residual subnet, specifically construct a residual block for residual learning, and activate it through a ReLU activation function to obtain a deep residual subnet prediction output; Step S34: constructing a gradient descent operator, specifically, updating the learning rate and loss function of the model through iterative calculation, performing iterative optimization of the model parameters of the artificial neural subnet, the convolutional neural subnet and the deep residual subnet, and obtaining the optimal model parameters; Step S35: subnetwork model integration, specifically integrating model output values ​​by weighted summation to obtain integrated model prediction data; Step S36: Cost anomaly detection model training, specifically, by constructing the artificial neural subnet, the convolutional neural subnet, the deep residual subnet, the gradient descent operator and the subnet model integration, model training is performed to obtain the cost anomaly detection model Model CD ; Step S37: Cost anomaly detection, specifically using the cost anomaly detection model Model CD , based on the financial system anomaly identification optimization data and the financial system anomaly identification feature data, cost anomaly detection is performed to obtain financial system cost anomaly feedback data.

4. The method for identifying abnormalities in an enterprise financial system based on artificial intelligence according to claim 3 is characterized in that: In step S4, the hybrid model combining the deep neural network and the convolutional long short-term memory network specifically includes a convolutional layer, a deep neural layer and a long short-term memory layer; The convolutional layer is used to extract local spatial features in the input data and serves as a hidden layer of the deep neural layer; The deep neural layer is used to construct the basic input-output structure of the hybrid model; The long short-term memory layer is used to extract the temporal features in the input data and serves as a hidden layer of the deep neural layer.

5. The method for identifying abnormalities in an enterprise financial system based on artificial intelligence according to claim 4 is characterized in that: In step S4, the step of using a hybrid model combining a deep neural network and a convolutional long short-term memory network to perform transaction anomaly detection and obtain transaction anomaly detection data includes: Step S41: construct a convolutional layer; Step S42: constructing a deep neural layer; Step S43: constructing a long short-term memory layer; Step S44: model hybrid connection; Step S45: Transaction anomaly detection model training; Step S46: Transaction anomaly detection.

6. The method for identifying abnormalities in an enterprise financial system based on artificial intelligence according to claim 5, characterized in that: In step S41, the convolution layer is constructed, specifically, a basic convolution layer including convolution weights and bias items is constructed, and a nonlinear activation function ReLU is used for activation to obtain a convolution feature output, and the convolution layer is constructed as the first hidden layer of the deep neural layer; In step S42, the deep neural layer is constructed, specifically, a deep neural layer including an input layer, a first hidden layer, a second hidden layer and an output layer is constructed, and the two hidden layers are activated using a ReLU activation function to obtain a prediction output of the deep neural layer; In step S4, the constructing of the long short-term memory layer specifically includes constructing a long short-term memory layer including a forget gate, an input gate, an output gate, a candidate cell state and a hidden state, and constructing the long short-term memory layer as a second hidden layer of the deep neural layer; In step S44, the model is hybridly connected, specifically, the first hidden layer of the deep neural layer is replaced by a convolutional layer, and the second hidden layer is replaced by a long short-term memory layer, and the model is hybridly connected to construct a hybrid model combining a deep neural network and a convolutional long short-term memory network; In step S45, the transaction anomaly detection model training is specifically performed by constructing the convolution layer, the deep neural layer, the long short-term memory layer and the model hybrid connection to obtain the transaction anomaly detection model Model ED ; In step S46, the transaction anomaly detection is specifically performed using the transaction anomaly detection model Model ED , based on the financial system cost anomaly feedback data and the financial system anomaly identification optimization data, transaction legacy detection is performed to obtain transaction anomaly detection data.

7. The method for identifying abnormalities in an enterprise financial system based on artificial intelligence according to claim 6, characterized in that: In step S5, the enterprise financial system anomaly identification is used to integrate the anomaly detection results of costs and transactions to identify the overall anomaly of the enterprise financial system, specifically by adopting the weighted average method to fuse the transaction anomaly detection data and the financial system cost anomaly feedback data, and by assigning weights to each detection result, calculating the comprehensive anomaly score, and setting the anomaly detection threshold according to the comprehensive anomaly score, and in the real-time monitoring of the enterprise financial system, when cost data or transaction data exceeding the anomaly threshold is monitored, an anomaly detection report is generated to obtain the enterprise financial system anomaly identification reference data; The enterprise financial system anomaly identification reference data specifically includes anomaly descriptions and anomaly correction suggestions.

8. An enterprise financial system anomaly recognition system based on artificial intelligence, used to implement an enterprise financial system anomaly recognition method based on artificial intelligence as claimed in any one of claims 1 to 7, characterized in that: It includes a data input module, a data processing module, a cost anomaly detection module, a transaction anomaly detection module and an enterprise financial system anomaly identification module.

9. The enterprise financial system anomaly identification system based on artificial intelligence according to claim 8 is characterized by: The data input module is used for data input, and obtains the original data of abnormality identification of the financial system through data input, and sends the original data of abnormality identification of the financial system to the data processing module; The data processing module is used for data processing, and obtains financial system anomaly identification optimization data and financial system anomaly identification feature data through data processing, and sends the financial system anomaly identification optimization data and financial system anomaly identification feature data to the cost anomaly detection module and the transaction anomaly detection module; The cost anomaly detection module is used for cost anomaly detection, obtains financial system cost anomaly feedback data through cost anomaly detection, and sends the financial system cost anomaly feedback data to the transaction anomaly detection module and the enterprise financial system anomaly identification module; The transaction anomaly detection module is used for transaction anomaly detection, obtains transaction anomaly detection data through transaction anomaly detection, and sends the transaction anomaly detection data to the enterprise financial system anomaly identification module; The enterprise financial system anomaly identification module is used for enterprise financial system anomaly identification, and obtains enterprise financial system anomaly identification reference data through enterprise financial system anomaly identification.

Images